Large language models are like Elmer Fudd, except they don’t need a bunny to confuse them, they do it themselves